OpenAI's superalignment team, co-led by @ilyasut
, has revealed its first research, exploring promising pathways to weak-to-strong model alignment (aka ways for puny humans to persuade ridonkulously smart AIs to obey them):

→ View original post on X — @willknight,